Introduction to Detectron2 Discussions

[ppwwyyxx]

Welcome to Detectron2 Discussions! This forum is a great place to ask the community "how-to" type of questions or discuss other interesting topics related to the project.

Similar to how projects like Pytorch, we will now use "issues" only for topics that potentially lead to commits: bugs/errors, feature requests, documentation improvements, etc. Topics outside the issue categories will be moved or directed to "discussions" instead.

"Discussions" is meant to be a community like Stackoverflow or PyTorch forum: asking clear, concise questions help you get answered. Some tips on using this forum more effectively:

• Stackoverflow provides some great guidelines on how to ask good questions
• Since we traditionally handle all questions in "issues", searching for answers in both places first might help
• Check our online documentation
• Human language is often ambiguous. If possible, use code/logs/screenshots/pseudo-code to describe your question

At last, please follow github community guidelines and our code of conduct in discussions.

Thanks for your interest in detectron2!

[eddydw]

Hi, How can I trained faster_rcnn_resnet5-_fpn with 3x224x224 images.
WHich parameters should be modified in which files?

[pkhateri]

Follow the steps in this notebook, and modify this function get_balloon_dicts(img_dir) for your own data.

[eddydw]

Thank you for your answer. Kind regards.

[kalikhademi]

Hi

I am trying to train a detectron model on BDD dataset and I got following error:
Traceback (most recent call last):
File "/home/.local/lib/python3.10/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap
fn(i, *args)
File "/second_study_codebase/detectron2/detectron2/engine/launch.py", line 123, in _distributed_worker
main_func(*args)
File "/orange/juangilbert/kalikhademi/dissertation_vision/dissertation_codebase/second_study_codebase/detectron2/downstream/object_detection/train_object_detection.py", line 93, in main
return trainer.train()
File "/second_study_codebase/detectron2/detectron2/engine/defaults.py", line 484, in train
super().train(self.start_iter, self.max_iter)
File "/orange/juangilbert/kalikhademi/dissertation_vision/dissertation_codebase/second_study_codebase/detectron2/detectron2/engine/train_loop.py", line 155, in train
self.run_step()
File "/second_study_codebase/detectron2/detectron2/engine/defaults.py", line 494, in run_step
self._trainer.run_step()
File "/second_study_codebase/detectron2/detectron2/engine/train_loop.py", line 310, in run_step
loss_dict = self.model(data)
File "/home/local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/home/.local/lib/python3.10/site-packages/torch/nn/parallel/distributed.py", line 1156, in forward
output = self._run_ddp_forward(*inputs, **kwargs)
File "/home/.local/lib/python3.10/site-packages/torch/nn/parallel/distributed.py", line 1110, in _run_ddp_forward
return module_to_run(*inputs[0], **kwargs[0]) # type: ignore[index]
File "/home/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/second_study_codebase/detectron2/detectron2/modeling/meta_arch/rcnn.py", line 161, in forward
proposals, proposal_losses = self.proposal_generator(images, features, gt_instances)
File "/home/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in call_impl
return forward_call(*args, **kwargs)
File "/second_study_codebase/detectron2/detectron2/modeling/proposal_generator/rpn.py", line 471, in forward
gt_labels, gt_boxes = self.label_and_sample_anchors(anchors, gt_instances)
File "/home/.local/lib/python3.10/site-packages/torch/utils/contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "/second_study_codebase/detectron2/detectron2/modeling/proposal_generator/rpn.py", line 340, in label_and_sample_anchors
matched_idxs, gt_labels_i = retry_if_cuda_oom(self.anchor_matcher)(match_quality_matrix)
File "/second_study_codebase/detectron2/detectron2/utils/memory.py", line 70, in wrapped
return func(*args, **kwargs)
File "/second_study_codebase/detectron2/detectron2/modeling/matcher.py", line 102, in call
self.set_low_quality_matches(match_labels, match_quality_matrix)
File "/second_study_codebase/detectron2/detectron2/modeling/matcher.py", line 121, in set_low_quality_matches
_, pred_inds_with_highest_quality = nonzero_tuple(
File "/second_study_codebase/detectron2/detectron2/layers/wrappers.py", line 153, in nonzero_tuple
return x.nonzero(as_tuple=True)
RuntimeError: nonzero is not supported for tensors with more than INT_MAX elements, file a support request

I know nonzero method is used in the wrapper.py but it gives me this error even for smallest batch sizes. So I wonder how can I solve this? can I use any other method in place of nonzero()?

[sulaiman1988]

how can I convert any pytorch model into ONNX